Spatial and Temporal Determinants of

Anthropogenic Forest Fires in the Amazon:

implications for forest degradation and future

carbon emissions

Ane Alencar, Gregory Asner,

Daniel Zarin, Francis Putz

• Forest fires were rare and mostly driven by extreme drought events

• Ignition sources were mostly natural and the forest was more resistant to fire

In the past

Today• Forest fires are becoming more

frequent, larger and perhaps mostly

driven by anthropogenic changes in

landscape than climatic events

• Most of forest fires are anthropogenic,

caused by escaped fires from human

activities

Understanding forest fire

(surface or understory fires)

1. Tracking forest fire history (Learn from past fire events)

2. Understand changes in fire regime (Spatial and temporal)

3. Estimate regional behavior of forest fires (Build the relationship with climate, land use, landscape structure, etc)

Study sitesDense

Open

Transitional

Iso-Data(clouds and defor.

Mapping)

CLAS-BURN

(PV-NPV)-Shade)

(PV-NPV)+Shade)

Overlay

fire scar

thresholds and filtering

Landsat

bands

PV, NPV,

ShadeReflectance

bands

Forest

maskBSI image

Masked

BSI image

Final

fire scar map

Mapping landscape

forest burn scars

Reference:* Asner, G. P., M. Keller, R. Pereira, J. C. Zweede, and

J. N. M. Silva. 2004. Canopy damage and recovery following

selective logging in an Amazon forest: Integrating field and satellite

studies. Ecological Applications 14:280-298.

•Development of a new index called Burn Scar Index (BSI)

•This index was based on an automatic calibration and sub-pixel analysis routine called CLAS-BURN, based on CLAS.

•CLAS* stands for Carnegie LandsatAnalysis System developed by the Asner Lab

Part 1

R:SB, G:PV, B:NPV Burn Scar Index (BSI) Burn Scar map

photosynthetic vegetation (PV) non-photosynthetic veg. (NPV) Shade/Burn (SB)

Burned forest accuracy: Dense forest 0.89; Open forest 0.79; Transitional forest 0.88

Burn Scar Index for unburned forest

and old and recent burns burned

Burn Extent

By forest type

Forest area

(ha)*

Average

annual

burned

area

Average

annual

deforestation

(ha)

% of total

forest

area

burned

% of forest

area

deforested

Dense 2,274,133 19,932 29,393 15% 29%

Open 2,324,883 104,711 62,821 44% 54%

Transitional 1,369,228 80,189 27,901 41% 50%

Changing fire regimes

Part 2

Fire sizes

Forest types have distinct fire

behavior in terms of size and total

area burned:

Majority of Dense forest fires scars

are small (< 100 ha)

In contrast to transitional forest where

most of the fire scars are large

Large fires also burned more area in

open and transitional forests

Fires between 100- 1000 ha in size

burn in average about the same area

each year in all forest types

Fire frequency

Most of the area burned was affected but 1

fire during the period

Fire interval

For the area that burned more than two times:

Fire interval for the dense forest appear

to be every 5 to 6 years, coincident with

ENSO

Fire interval for transitional forest have

a higher return after 2 or 3 years, fuel

limitation

Dense

Open

Transitional

Fire SeasonDense

Open

Transitional

Fire season in the 3

regions are getting about

one month late

Fire intensity and effectsRelationship between frequency and canopy cover

Impacts of burn frequency in forest

structure

0

3

Burn frequency

*Explains 65% of variation

Fate of burned forest

Forest area

(ha)*

% of forest

area

deforested

% of total area

deforested that

was burned

% of total

burned area

that was

deforested

Dense 2,274,133 29% 19% 38%

Open 2,324,883 54% 39% 48%

Transitional 1,369,228 50% 38% 46%

19 to 38% of the deforested area was burned

38 to 48% of the burned area was deforested

Part 3

Relatioship with forest

clearings

Fires penetrate deeper in Transitional

forests than the other forest types

90% of the area burned is within 5 km from

a clearing

The highest frequencies also happen within

1 km from a forest edge

Dense

Open

Transitional

Extrapolating results and creating

fire probabilities for wet, average and dry years

Forest fires

in 3 distinct

climate

conditions

Fire probabilities based

on PAW and

fragmentation for wet,

average and dry years Some of the areas (blue circle) already

showing influence of fragmentation in

changing the likelihood of fire in average

rainfall years, and even in wet years.

These areas are believed to have reached

the tipping point where fragmentation has

played a more important role to the forest

fire occurrence than climate.

The dry years fire probability map indicate

the areas under higher risk of forest fires,

where forest is flammable due to extreme

drought and high ignition sources

probabilities.

Estimated commited CO2 emissions

from deforestation and forest fires

for the three forest types during the

last 24 years

Forest types

Deforestation

-driven CO2

emissions1

(Tg yr-1)

Forest fire-driven committed CO2 emissions2

(Tg yr-1)

Average

annual area

burned Wet years

Average

years Dry years

Dense 17.6 6.0 0.1 2.4 14.7

Open 27.6 23.0 3.7 15.7 48.8

Transition 13.8 19.9 3.6 12.2 40.9

59.0 48.8 7.4 30.2 104.3

1 The CO2 emissions for each forest type were calculated using the Saatchi et al. (2007) biomass map, in which the average biomass value for each vegetation type was converted to Carbon and multiplied by the annual area deforested, and then converted to CO2.2 The committed CO2 emissions from forest fires was based on the average tree mortality due to forest fires reported on literature (Alencar et al 2006), not including yet the released emissions during the fire itself.

Estimated area at risk of burning,

area burned and CO2 emissions by

forest type and climatic conditions

for the Brazilian Amazon

Forest

Type

Area total by

forest type

(thousand

km2)

Estimated area

at risk of burning

(thousand km2)

Estimated area burned

(thousand km2)

Estimated emission

(Pg CO2 yr-1)

WET AVE DRY WET AVE DRY WET AVE DRY

Dense 1,783.8 2.4 21.3 160.3 0.3 2.2 6.1 0.01 0.07 0.18

Open 884.5 4.5 35.6 121.3 1.6 10.2 22.3 0.04 0.22 0.49

Transitional 504.7 3.1 13.5 81.8 1.6 5.5 23.5 0.04 0.14 0.58

Total 3,172.9 10.0 70.4 363.5 3.5 18.0 51.9 0.08 0.43 1.25

The estimated area burned is the portion of the estimated area under risk

of burn that is located up to 5km from forest clearings

The estimated area burned can be approximately the same in average rainfall years than the average annual deforestation rates during the last decade.

• (1) severe droughts are the main temporal

determinant of forest fires having overall emissions

that were 76% higher than average deforestation emissions;

• (2) although, since these are not wildfires but escaped fires from

anthropogenic land use sources, the spatial distribution of these

fires revealed a pattern where ~90% of the area burned occurs

within 10 km of official roads, 1-2km of deforested clearings,

and within highly fragmented areas;

• (3) the spatial and temporal characteristics of ENSO fires

disproportionate impact dense forests;

• (4) escaped forest fire emissions are historically large, especially

in ENSO years, and growing in importance as deforestation

emissions drop and escaped fire emissions increase, associated

with increasing importance of small slash-and-burn clearings

Main results

Thank you

Dan Zarin, Jack Putz, Greg Asner, WendellCropper,Charles Wood,

Daniel Nepstad, Paulo Brando, Jennifer Balch and Claudia Stickler, Mike Coe

Academic Support Funding supportNSF – DDRI

NSF DEB-0410315 NASA NESSF Program

Tropical Conservation and DevelopmentProgram -TCD

Compton FoundationAmazon Conservation Leadership Initiative –

ACLIMoore Foundation

Florida-Brazil Program